Keywords: energy performance index (EPI), residential buildings, Monte Carlo analysis, neural networks, approximation models, design optimisation, variable pruning, performance uncertainty.

Accurate estimation of the energy consumption of buildings (taking into account all significant dynamic environmental factors, including the behaviour of building users) can be involved, requiring a lot of time and resources. Researchers have developed several approximation models over the years that attempt to estimate energy consumption from a few parameters, including the Energy Performance Index (EPI) of the building [1]. EPI is deterministic in that it assumes the values of the independent variables are known precisely. In reality, these values are subject to variance and uncertainty. Knowledge of the amount of uncertainty in the EPI and energy consumption for a building is useful for many reasons, including determining the probability of alternative designs actually meeting the Florida code. In this paper, the authors describe a method of estimating the variance as well as the expected energy consumption of a residential building using neural networks. The paper concludes with a brief proposal for future research in this area.

This paper reports on recent developments of an on-going research project concerned with estimating the energy consumption and Energy Performance Index (EPI) of residential buildings. The work on which this study is based [2] has been successful in linking the most influential of the approximately 100 factors used to calculate EPI to the actual EPI score. In particular, this work attempts to create a model that links EPI to actual energy consumption, and proposes a method of incorporating uncertainty into the estimate. Such developments will increase the usefulness of the EPI approximation technique to designers of residential buildings.

As with earlier research, this model was created using a back-propagation artificial neural network. Artificial neural networks have the ability to link given inputs and outputs by finding the correlation between them. Based on this, the neural network creates a pattern, or learns, from the data. Once it has learned how inputs and outputs are related, it can take a given input and return the correct output. In other words, neural networks have the ability to correctly classify or predict an outcome based on inputs it has not learned or "seen" before. Ultimately, the intent is to develop the system to work for situations where there is significant uncertainty and variance in the values of the inputs (such as the thickness of insulation, the actual solar loading, and the occupant usage/behaviour), and to predict from this the expected energy consumption and the corresponding variance/uncertainty of this value.

The paper concludes with a brief proposal for future research in this area.

1

Florida Model Energy Efficiency Code for Building Construction, Laws of Florida Chapter 81-226 and Florida Statutes Chapter 553, Part VII, 1998.

2

Issa, R. R., Flood, I., and de Martini, A., "Estimating the Energy Performance Index of Buildings" in "Artificial Neural Networks For Civil Engineers", I. Flood and N. Kartam (Editors), ASCE, Reston, VA, 260-272, 1998.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £85 +P&P)